Literacy Sharing, Assortative Mating, or What? Labour Market Advantages and Proximate Illiteracy Revisited

Abstract

This paper explores the relationship between household literacy and the labour market outcomes of illiterate household members which Basu, Narayan and Ravallion (2002) report using Household Income and Expenditure data from Bangladesh. BNR attribute a considerable wage premium for proximate-illiterate women in off-farm employment to labour productivity gains from intra-household literacy sharing. This wage premium also suggests that women may be more efficient recipients of literacy externalities than men. We propose that any such relationship might not be due to higher labour productivity but may have other explanations such as systematically different and unobserved attributes of illiterate females married into literate households. We also pay attention to the negative selection of illiterate females into non-farm wage employment, which contrary to received wisdom suggests that household literacy may not be unambiguously progressive for females. We propose that the widely reported finding that female literacy impacts more positively than male literacy on child wellbeing may not extend into similar effects in other realms of household activities where males may be more efficient transmitters of literacy externalities. Using more recent Bangladesh and similar Indian data we find somewhat different results for household literacy externalities on non-farm wage employment of household illiterates, and also show that any such effects are conditioned on the social identity of the individuals, their geographic location and their sector of employment. We caution against drawing conclusions from one finding using one data set apparently ignoring contrary findings, where that finding is congruent with fashionable development views, such as the advantages of females as generators of development.

Notes

1. We are happy to acknowledge, without implicating, discussions with Kausik Basu and Bryan Maddox, and the comments of a reviewer.

2. Such high estimates are increasingly contested because of their likely upward bias (for example Behrman and Wolfe, 1984). For instance, using data from Lucknow, Uttar Pradesh, Kingdon (1998) shows that estimates of educational returns are highly sensitive to the omission of variables capturing school quality variation.

3. The terminology is not always easy to follow; female proximate illiteracy, or f-proximate illiteracy means that the illiterate person (female or male) lives in a household with female literates, whereas female proximate illiteracy means that the illiterate person is female and lives in a household with at lest one literate (who may be female or male).

4. As noted above, others using these ideas have focused on the association of proximate illiteracy to health outcomes or behaviours (Gibson, 2001; Alderman et al., 2003). It should be noted that while Basu et al. (2002) use household level identifiers of the literacy characteristics of households (that is, a household has literate females, and/or males), these other authors use community level variables (that is the proportion of female literates in the community).

5. It appears that some authors designate households as m- and f-proximate in asymmetric ways which may bias their results in favour of larger effects of f-proximity. Thus, both Basu and Foster (1998) and Gibson (2001) identify m-proximate households as ones with only male literates and no female literates (m-only literate households), but identify ‘an f-proximate illiterate [household as one that] … contains at least one female literate’ (Basu and Foster, 1998: 1745; Gibson, 2001: 162). This appears to imply that m-proximate households have only male literates (m-only literate), f-proximate households include both female-only literate households (f-proximate) and households with both male and female literates (m&f-proximate). Given the higher proportion of male literates it is likely that there will be more m-proximate than f-proximate illiterates, but it is also likely that the Basu and Foster (1998) and Gibson (2001) definitions result in all m&f-proximate illiterates being assigned to the f- proximate illiterates category. Since the literacy externalities of the latter (m&f-proximate) are likely to be greater (Dutta, 2004) this will inflate the literacy externalities received by the f-proximate category. See the results section where we find some support for the idea in that the coefficient of the m&f-proximate illiteracy is greater than either the m- or f-proximate variables.

6. Alaka Basu goes on to suggest that some or all of the association of maternal literacy with fertility decline may consequently be attributed as much to the characteristics of the males whom literate females marry, as to the literacy of the females per se.

7. As noted by Dougherty (2006), wage equations based on cross-sectional data often find an earnings premium of around 10 per cent for married men.

8. Basu et al. (2002) argue that it is not implausible that literate men may be more willing to share their literacy with illiterate females than with illiterate males because the females are more likely to spend any increased income on children (whose welfare features in the utility functions of both females and literate males to a greater extent than in those of illiterate men). Apart from the possible contradiction (why are children in the utility function of literate men but not in that of illiterate men?), this requires testing the literacy sharing effect on male and female proximate illiterates of male-proximate illiteracy. This can be done by running regressions for male-only literate households, which Basu et al. (2002) do not do.

9. A data appendix discussing the comparability of the data from these surveys is available: http://139.222.138.83/literacy/Data Appendix.pdf

10. A further point to note is that while the Mincerian specification uses hourly wages as the dependent variable, Basu et al. (2002) use earnings, which may confound wage rates (productivity) with the allocation of time between market labour participation and leisure or home work (Killingworth and Heckman, 1986).

11. Kingdon (1998) finds that the rate of return to male education falls by nearly 20 per cent when a cognitive score test is used rather than paternal education as the variable representing the quality of the learning environment.

12. MLE in Stata 6; we have not always been able to reproduce the Basu et al. (2002) results using ostensibly the same data set and version of Stata, although broadly we get the same results (see Appendix tables). It is not clear whether the differences we experience are due to differences in the data provided by the BBS (a not uncommon phenomenon), reporting errors, or minor differences in data processing and estimation. We use Heckman ML with cluster ( ) and robust options for HIES 1995/1996 except where it did not converge, but for other data sets we have used the twostep procedure. In general all coefficient signs and sizes are similar between the two procedures.

13. Marriage is thus patrilocal, hypergamous and exogamous.

14. Rahman and Rao (2004) contend that the North-South cultural divide is no longer a good characterisation of differences in marriage patterns or their effects on female and child welfare, and report notable findings such as the lack of significance of difference in distance between natal and conjugal residence. However, Rahman and Rao (2004) is based on a data set (sample survey) with limited geographical coverage compared to that on which previous characterisations have been based (Indian census). This claim needs further empirical investigation, especially as it is not consistent with results from the near contemporaneous National Family Health Survey, 1998.

15. Though one may note that their sample includes both never married, and widowed and divorced females which substantially alters the interpretation of their findings.

16. We find 688 illiterate females with non-agricultural wage employment, and 601 and 375 illiterate males in the rural and urban sectors respectively with non-agricultural wages (see Appendix Table 1). We have no explanation for these discrepancies.

17. http://139.222.138.83/literacy/Literacy Tables.doc

18. As noted in the data appendix (available from http://www.139.222.138.83/literacy/ProximateLiteracy.html), the other surveys do not allow use of exactly the same variables due to differences in the variables reported.

19. Note that this was estimated by the twostep procedure because ML did not converge. When the model is estimated for never-married by twostep, the coefficient in the wage equation is 0.495. When pooled with dummies for marital status interacted with household literacy, it appeared that there was no significant difference in the effect of household literacy on non-farm earnings of illiterate women.

20. Unfortunately, we cannot exclude the effect of rising female literacy which will tend to have raised the literacy of younger wives, or of other unobserved characteristics of wives of different marital experience.

21. Proxied by age; we have no data directly on marital experience.

22. A total of 89 per cent of illiterate females are male-proximate including those in households where both males and females are literate.

23. When we use what we assume is the Basu et al. specification (Basu et al., 2002: 661), we find that the coefficients of m- and f-proximity are significant and negative for both male and female illiterates, and, although not significantly different from from each other, the coefficients of f-proximity are more negative than those on m-proximity, again indicating that, if anything, male literates are more effective transmitters of wage premium literacy externalities than female literates.

24. The failure to explore the externalities of male attributes is so embedded in analytical practice that in many cases the relevant attributes of males are neglected. Typical examples of this are the failure to record the nutritional or health status of fathers (or other adult males) in surveys of nutritional wellbeing. Demographic and Health Surveys that are increasingly available and used to assess levels and progress in health and nutrition indicators of wellbeing (http://www.measuredhs.com/aboutdhs/whoweare.cfm; Deaton, 2003) provide one example. This lacuna is even more pronounced in the Bangladesh Child Nutrition Surveys which we used to assess spillovers to child wellbeing from the wage earnings advantages of female proximate illiterates.

25. The effects of proximate illiteracy on child health have been examined by Gibson (2001) and by Alderman et al. (2003), using community level literacy variables. This is clearly problematic. In our models we use a dummy variable for whether the household has a literate member.

26. We do not address the possibly confounding effects of relationships between the education of both parents and child welfare in this paper.

27. The Quinquennial Rounds have in recent decades involved large samples in parallel Consumer Expenditure and Employment and Unemployment Surveys. The annual surveys between the quinquennial rounds have smaller samples and focus on a diversity of topics and, hence, do not provide such reliable statistics. We report here only the 38th and 43rd round results; 50th round results are available from the authors.

28. Although it could, in part, especially for STs who are concentrated in backward regions of India be due to lesser development on the non-farm economy in these areas.

29. Bangladesh would generally be included in the Northern cultural regime.

30. HIES, 2000/2001 also allows wage as well as earnings equations, although these are not reported here.

31. As noted above, complete tables can be obtained from the authors on request.

32. Results for the 50th round of the EUS, not reported here, showed other differences with the two preceding Quinquennial rounds.

33. This could of course be argued to be due to their low bargaining power, especially for widows (Chen and Dreze, 1992); however, it is also not inconsistent with their experiencing more literacy externalities over the long period they have been in the household, or with their original selection into the literate household on the basis of having greater ability than other illiterate females notwithstanding their lack of education.

34. We do not consider the possibility that it is due to errors in measurement of literacy, although this could be a factor (Griliches, 1977; Krishnan, 1996).

35. See Leamer's (1983) classic critique of econometrics and the contributions of Kennedy and others to the discussion of applied econometrics for sundry cautions (Kennedy, 2002).